Am Donnerstag, 24. November 2016, 17:14:56 schrieb Alexander Kochetkov: > > 24 ????. 2016 ?., ? 16:21, Heiko St?bner <heiko at sntech.de> ???????(?): > > > > what I actually meant was that the driver could also recognize the rk3188- > > timer compatible as "we need a clocksource" and it shouldn't matter which > > timer actually gets used for this. > > One rockchip timer cannot be used as clockevent and clocksource at the same > time. > > In case of clockevent we want interrupts from it at specified times. So we > load one value into timer counter and it generates an interrupt. > > In case of clocksource we load max value into timer counter, run timer and > read current value on demand. > > rockchip_timer driver currently implement clockevent. So, if I create only > one timer in the device tree, it should be clockevent timer. As that > behavior already expected from driver by people used it. > > I may suggest such solution here: if I want clocksource, I have to declare > two timer in device tree. First probed timer would be clockevent and second > one would be clocksource. All other timers will be ignored. Is that > solution good? yep, sounds good, especially as with your patch 9/9 you already declare these necessary timers. > If I want one timer and want it be clocksource not clockevent how that > situation should be configured? Device tree not good for this. Kconfig not > good. Pass that configuration on kernel command line? simply ignore that case :-) I.e. newer kernels are supposed to be able to run old devicetrees and in that case they will have the global-timer as (slightly unstable) clocksource. Also on the cortex-a9 we also still have the smp-twd as clockevent device. Heiko
